Omni Flash preview dies 30 Sept

gemini-omni-flash-preview dies on 30 September 2026. Google said so on 27 August 2026 in the Gemini API changelog, the same day it shipped gemini-omni-1.1-flash as generally available. As of 19 September 2026 that is eleven days. Scripts, notebooks, and wrappers that still pass the preview id will fail. The replacement string on Google's door is gemini-omni-1.1-flash. Versely does not publish /models/gemini-omni-1-1-flash. If you generate on this bill, you are on Gemini Omni Video or Gemini Omni Flash Edit. Those are catalog rows. They are not the preview endpoint.

This page is only the cutoff. The GA feature list is a different post. Forty-second extend is a different post. 360p drafts are a different post. Do not read this as a model review.

The preview string is the whole story

Two ids. One date. That is the news.

On 30 June 2026 Google released gemini-omni-flash-preview in public preview. The changelog called it a high-performance multimodal model for high-speed video generation and conversational video editing. Through the Interactions API you could generate 3 to 10 second videos at 720p from a text description or a still, then edit the output by talking to it. That paragraph is why the string is in so many repos. Cookbooks, Colabs, internal wrappers, and "we'll pin preview until GA" configs all copied it.

On 27 August 2026 Google released gemini-omni-1.1-flash as the GA version of that fast conversational video model. Same changelog entry, same day: "The existing gemini-omni-flash-preview endpoint will be deprecated on September 30, 2026." The Keyword post that morning is the marketing surface for the GA id. Its Python sample already uses model="gemini-omni-1.1-flash". It does not use the preview string. If your repo still does, you are on the June sample, not the August sample.

Google's deprecations table (last updated 17 September 2026 UTC) lists the same pair under Gemini Omni Flash models:

• gemini-omni-1.1-flash, released 27 August 2026, no shutdown date announced
• gemini-omni-flash-preview, released 30 June 2026, shutdown date 30 September 2026, recommended replacement gemini-omni-1.1-flash

Google's footnote on that page is real, so quote it instead of pretending it isn't there. Shutdown dates in the table are the earliest possible dates a model might be retired, and Google says it will communicate the exact shutdown with advance notice. Plan against the changelog sentence anyway. Treat 30 September 2026 as the day the preview string stops being a thing you can call. If Google later slides the hour, you have not been harmed by switching early. If it does not, a queue of jobs with the old id is the outage.

Preview is not a nickname. It is a literal suffix on a model id. gemini-omni-flash-preview and gemini-omni-1.1-flash are different endpoints. A wrapper that concatenates gemini-omni-flash plus a hard-coded -preview is the bug. A .env that still says GEMINI_OMNI_MODEL=gemini-omni-flash-preview is the bug. A CI fixture that asserts the request body contains that string will go red for the right reason, once you change it.

The Gemini Omni Flash model card, last updated 27 August 2026, still lists both under Versions: stable gemini-omni-1.1-flash, preview gemini-omni-flash-preview. That is documentation lag, not a stay of execution. The cutoff documents are the changelog and the deprecations table. Do not keep the June id because a versions list has not been trimmed yet.

This is not a quality argument. GA is not "the preview but nicer." GA is the supported id. Preview is the id Google has dated.

Grep it before 30 September

You have eleven days from 19 September 2026. That is not a quarter. It is a grep, a PR, and a deploy.

Search the whole org, not the one service you remember shipping in July:

• gemini-omni-flash-preview as a string literal in application code
• omni-flash-preview in case someone stripped the gemini- prefix
• YAML, JSON, TOML, and Terraform that inject the model name
• LangChain, LlamaIndex, n8n, Zapier, and Make modules that stored a June dropdown
• Colab notebooks in Drive, not only the Git repo
• Cron, Cloud Scheduler, and GitHub Actions payloads
• Retry queues and dead-letter messages that serialized the original request, model id included
• Eval harnesses, golden files, and snapshot tests
• Internal docs and runbooks that still say "use the Omni Flash preview model"

Change the string. Do not wait to see if it still 200s on 1 October. Google's deprecations page is explicit about the end state: once a model is shutdown, it is completely turned off, and the endpoint is no longer available. Planning for a soft 404 is how a Monday batch misses the week.

Do not invent a new Versely slug while you are in there. There is no /models/gemini-omni-1-1-flash in the catalog as of 19 September 2026. Confirm it yourself. The published Omni generate row is Gemini Omni Video: minimum 29 credits; durations 4s, 6s, 8s, and 10s; 720p / 1080p / 4K; 16:9 or 9:16. The audio field on that card is null. Features include multimodal, text-to-video, image-to-video, reference-to-video, character-ref, audio-driven, and 4k. Audio-driven is a feature flag. It is not a claim of native audio. The published edit row is Gemini Omni Flash Edit: minimum 50 credits; audio false; feature edit_video; video in, instruction in, revised video out.

Those two rows are what a Versely credit buys. They are not the Gemini API preview endpoint. A Versely job does not send gemini-omni-flash-preview because you picked Omni Video. A script you wrote against Google AI Studio in July does.

Who this deadline hits:

• Anyone calling the Gemini API with model set to gemini-omni-flash-preview
• Sample apps copied from the 30 June cookbook
• Vendor wrappers that pinned the preview id "until GA"
• Staging environments that nobody rotated after 27 August
• Eval jobs that still assert the June id in the request body

Who this deadline does not hit:

• A Versely user generating on Omni Video or editing on Omni Flash Edit
• A Gemini app or Flow user already on 1.1 after 27 August. Google's Keyword post put Omni 1.1 in Flow for AI Plus, Pro, and Ultra, and scene extension in the Gemini app
• A Veo job. This is not a Veo model id. Do not "fix" a Veo call because Omni preview is dying

If you are not sure which bucket you are in, grep. The string is the test. A producer who only presses generate on Versely can stop reading after this paragraph. An engineer who wrapped Gemini in June cannot.

Thirty-four days of warning ran from 27 August to 30 September. As of 19 September, eleven remain. Preview lived ninety-two days, 30 June to 30 September. Teams that treat preview ids as production ids get this calendar every cycle. The work this week is a string replace. It is not a rewrite of the Omni brief.

The replacement is gemini-omni-1.1-flash

The replacement Google named is gemini-omni-1.1-flash. Changelog, 27 August 2026. Deprecations table, recommended replacement, same id. Keyword post Python sample, same id.

That is a Google API string. You put it in a Gemini API model field, an AI Studio prompt, or an Agent Platform call. You do not put it in a Versely catalog path. /models/gemini-omni-1-1-flash is not a page we publish. Do not bookmark a 404 and call it a migration.

If the job is "I call Google myself":

1. Change the model id to gemini-omni-1.1-flash.
2. Re-read the Omni guide and the 1.1 samples. Request shape may have grown (extend task, resolution in video_config, first-last via image_to_video with up to two images). Those controls are GA's. They are not this page. Omni 1.1 Flash: what changed owns the control list. This page owns the string.
3. Run one generate and one edit against a throwaway key before you flip production.
4. Update the eval set so it expects the new id, including snapshots that hashed the old request body.
5. Leave a comment in the PR with the three dates: preview 30 June 2026, GA plus deprecation notice 27 August 2026, cutoff 30 September 2026.

If the job is "I generate Omni on Versely":

You already do not use the preview id. Stay on Gemini Omni Video for a 4 / 6 / 8 / 10 second multimodal generate, or Gemini Omni Flash Edit for a conversational retouch. Caption on the AI caption generator. The AI video generator is the Versely door for the rows we publish. Sign up if you needed that door instead of a dying preview string.

If the job is "I needed a 1.1-only control":

You are on Google's surface for that control. AI Studio, Agent Platform, Flow, Gemini app extend. Bring the file back. Do not wait for a Versely 1.1 slug to appear before you migrate the preview id. The preview id is dying whether or not we add a third Omni row.

Do not mix meters. Google's 1.1 pricing table lives on the Keyword post as a screenshot. Versely credits live on the catalog cards. 29 minimum on Omni Video is not a Google per-second dollar figure. 50 minimum on Flash Edit is not a Google invoice line. Quote each bill in its own unit.

Print the four names if you have to:

• Dead on 30 Sept 2026: gemini-omni-flash-preview
• Google's GA replacement: gemini-omni-1.1-flash
• Versely generate: gemini-omni-video, min 29 credits
• Versely edit: gemini-omni-flash-edit, min 50 credits

Four names. One of them is about to stop answering.

What this cutoff does not change

A model id sunset is a tempting moment to rewrite the whole Omni brief. Don't.

It does not change the Versely Omni generate lengths. Still 4s, 6s, 8s, 10s on Omni Video.

It does not add a forty-second Omni row here. Forty seconds is Google 1.1's cumulative extend cap on Google's surface. That argument is Omni 1.1 extends to 40 seconds, and it is not this deadline. Do not route that job through a Veo row because both logos say Google.

It does not make 360p drafts a Versely SKU. Google's 360p claim (up to 60% faster and one-third the cost of 1.1's 720p, 27 August 2026, system throughput) is Omni 1.1 360p drafts. Still not this string.

It does not turn Omni Video into native-audio Veo. Omni Video's audio field is null. audio-driven is in features. Flash Edit's audio is false.

It does not create /models/gemini-omni-1-1-flash.

It does not deprecate Gemini Omni Video or Gemini Omni Flash Edit. Those are our published rows as of 19 September 2026. Google sunsetting a preview endpoint is Google's changelog. It is not a Versely takedown.

It does not require you to leave Omni. It requires you to stop sending a string Google has dated.

June 30 to September 30 is ninety-two days of preview. The 27 August notice was thirty-four days of warning. As of 19 September you have eleven. That is enough if the work is a string replace. It is not enough if the work is "we'll migrate when 1.1 lands in our catalog." 1.1 already landed on Google's door. The catalog honesty is: we have Omni Video and Omni Flash Edit. We do not have a 1.1 slug. Those two facts can both be true on 30 September.

FAQ

Does Versely Omni Video die on 30 Sept?

No. Gemini Omni Video is a published catalog row (min 29 credits, 4 / 6 / 8 / 10s). Gemini Omni Flash Edit is a published catalog row (min 50 credits, edit_video, audio false). Google is shutting the Gemini API preview id gemini-omni-flash-preview. That is a different string on a different door.

Is gemini-omni-1.1-flash in the Versely catalog?

No. There is no /models/gemini-omni-1-1-flash. Google's GA id is gemini-omni-1.1-flash on AI Studio, the Gemini API, Agent Platform, Flow, and the Gemini app. Use that string when you call Google. Use Omni Video or Omni Flash Edit when you call Versely.

Is 30 September a maybe, or a hard stop?

Treat it as a hard stop. The 27 August changelog says the preview endpoint will be deprecated on 30 September 2026. The deprecations table lists that date as the shutdown date, replacement gemini-omni-1.1-flash. Google's footnote says table dates are earliest possible and that an exact shutdown can still be communicated. Do not use that footnote as permission to ship the old id into October. Switch the string this week.